Research

"LLM Meets Job Advertisements: Unmasking Skill Premiums in UK" (Job Market Paper) , June 2024, download

The rapid advancements in technology and events like the COVID-19 pandemic have significantly transformed the modern workplace, potentially altering the skills required for success. This study estimates the wage premium associated with Information and Communication Technology (ICT), interpersonal, and Artificial Intelligence (AI) skills within occupations in the UK. I leverage a novel machine-assisted mixed method utilising job advertisements from April 2016 to December 2022.  Skills are extracted from job descriptions and  categorised using zero-shot learning with the Large Language Model (LLM) - GPT-4 application programming interface (API). Throughout the observed period, interpersonal skills remained consistently high in demand, with around 90% of jobs requiring them. In contrast, ICT skills experienced a negative impact during the COVID pandemic but have since rebounded to a demand level of around 55%. The demand for AI skills has remained relatively stable, with approximately 13% of jobs requiring them. I find no significant wage premium for interpersonal skills, likely due to their widespread requirement  across occupations. However, ICT skills command a premium, ranging from 4% for basic ICT proficiency to 13% when combined with AI expertise. Notably, the premium for all three skill sets (ICT, AI, and interpersonal) doesn't statistically differ from the premium for ICT and AI alone. This research contributes in three ways. First, I introduce a novel LLM-based text classification approach for social science analysis. Second, I estimate skill premiums within occupations, providing a more nuanced understanding of their economic value. Third, I investigate the impact of the COVID-19 pandemic on skill premiums, informing policymakers on dramatic impact in the short run but little long effects on skill demand and premiums.

"Automation and Human Capital Investment" , April 2024, download

Technological change has an ambiguous impact on labour market by creating demand for some skills and reducing demand of some others. Hence the relationship of demand for skills and the need for training is an empirical question. In this paper, I investigate how different types of technological change affect the demand for training. I categorise several measures of automation on the basis of tasks  (done at individual or occupation level) and technology (Software, Robot, AI) and compare their effect on human capital investment. I find that the effect of automation on training is sensitive to both measure of automation used, reducing with individual and occupation level automation measures. Automation reduces the incidence of training. However when relying on technology based measures of automation, workers affected by older technologies (Robot, Software) receive less training with automation, while workers affected by newer technologies (AI) receive more training with automation. The findings are similar for workers of different age groups and skill levels.

"Standard Occupation Classifier - A Natural Language Processing Approach " (with Jack Patman) , March 2024, download

Standard Occupational Classifiers (SOC) are systems used to categorize and classify different types of jobs and occupations based on their similarities in terms of job duties, skills, and qualifications. Integrating these facets with Big Data from job advertisement offers the prospect to investigate labor demand that is specific to various occupations. This exploration encompasses critical aspects including wage structures, skill prerequisites, and geographical distribution. This project investigates the use of recent developments in natural language processing to construct a classifier capable of assigning an occupation code to a given job advertisement. We develop various classifiers  for both UK and US SOC, using different Language Models. We find that an ensemble model, which combines Google BERT and a Neural Network classifier while considering job title, description, and skills, achieved the highest prediction accuracy. Specifically, the ensemble model exhibited a classification accuracy of up to 61\% for the lower (or fourth) tier of SOC, and 72\% for the third tier of SOC. This model could provide up to date, accurate information on the evolution of the labour market using job advertisements.

"Using Topic Modelling to Discover New Trends in the Scientific Literature" (Data Study Group Final Report: Defence Science & Security Laboratory ) , May 2023, download

Alan Turing Institute Data Study Group (DSG) explored the use of topic  modelling  to  discover  new  trends  in  the  scientific  literature.   The challenge   was   presented   by   the   Defence   Science   and   Technology Laboratory  (Dstl)  who  have  been  working  in  this  area  for  a  number  of years.   The aim of this DSG is to advance the work already started by Dstl, whilst also building a better understanding of the specific research challenges that automated horizon scanning techniques present and how these  can  be  overcome  in  the  future.   As  the  science  and  technology landscape  continues  to  rapidly  shift and  evolve,  this  challenge  will  help policy makers to stay on top of these latest developments.